Possible to use AlignedUMAP with precomputed distance matrices

[jonasenander]

I have a bunch of precomputed distance matrices for time continous data and wanted to use AlignedUMAP on that dataset. Ergo, I create the slices and relationship data as is instructed. However, in the standard case you will have slices with the structure:
List[ Array with shape <window length, dimensionality of data> ].
Meaning that on line aligned_umap.py:352 the X[n] that is sent into the UMAP mapper is <window length, dimensionality>.

However, if you provide precomputed square distance matrices instead you will have slices with the structure:
List[ Array with shape <window length, number of data points, number of datapoints> ].
Thus giving you that X[n] is now a 3dimensional array. This flunks the check_array method on umap_.py:2372.

Am I doing it wrong? Any way around it? Does AlignedUMAP not support precomputed distance matrices?

[lmcinnes]

I believe this should be possible (but haven't verified there aren't lurking dragons). In principle for the regular case you should have a slice that is n_samples x n_features and for a precomputed metric you should simply have something where each slice is n_samples x n_samples so the overall number of axes should not be changing between the two. Perhaps I am not properly understanding your setup or what you are trying to do?